Revert 64-bit interlocked intrinsics on x86

[StephanTLavavej]

This ports MSVC-PR-269581, reverting GH-986 / MSVC-PR-240462.

According to John Morgan:

We recently discovered a conflict between the 64-bit Interlocked
intrinsic functions newly implemented for x86 and the Windows 8.1 SDK.
To resolve the conflict we are reverting the change. We may later
implement an alternative solution to support these functions on x86.

[CaseyCarter]

Should we remove #1018, or repurpose it to remove the new codepaths for Clang as well?

[StephanTLavavej]

We should probably tell Clang that we're reverting the change, so no action is requested from them at this time, if I understand this correctly.

[CaseyCarter]

We should probably tell Clang that we're reverting the change, so no action is requested from them at this time, if I understand this correctly.

I think I was trying to say "We should close #1018 and continue to use the same codepaths for Clang and MSVC instead of taking advantage of these intrinsics that we asked for which now will only exist on Clang." but apparently got lost along the way.

[AlexGuteniev]

Clang folks may need to revert their change as well -- for the same reason.

Honestly I'd prefer hacking around the SDK instead.

[StephanTLavavej]

I realized that Clang might not care about compatibility with the ancient Win8.1 SDK. However, I agree about #1018 and will go close it now.